def highFrequencyConv(img, tmpl): Q = metrics.slidingWindowH(img,inner=7,maxM=50,norm=True)[20:32, :] mf = cv2.matchTemplate(Q.astype('Float32'), tmpl, cv2.TM_CCOEFF_NORMED) print mf.shape return mf
import metrics import fileio import numpy from matplotlib import mlab, pyplot as plt import pandas df = pandas.DataFrame( numpy.zeros(30000 * 5901).reshape(30000, 5901), columns=(["label"] + map(lambda a: str(a), range(100 * 59))) ) labels = pandas.read_csv("../data/train.csv") for i in range(30000): img = fileio.ReadAIFF("../data/train/train{}.aiff".format(i + 1)) P, freqs, bins = mlab.specgram(img, NFFT=256, Fs=2000, noverlap=192) Q = metrics.slidingWindowV(P, inner=3, maxM=40) W = metrics.slidingWindowH(P, inner=3, outer=32, maxM=60) s = numpy.vstack((Q, W)) df.ix[i] = numpy.append(labels["label"][i], s.reshape(100 * 59)) if i % 100 == 0: print i df.to_csv("total.csv", index=False)
def slidingWindowH(img): w = metrics.slidingWindowH(img, inner=3, outer=32, maxM=60) print w.shape return w